{"id":383,"date":"2022-08-01T14:18:22","date_gmt":"2022-08-01T06:18:22","guid":{"rendered":"http:\/\/www.gislxz.top\/?p=383"},"modified":"2022-08-01T14:49:55","modified_gmt":"2022-08-01T06:49:55","slug":"%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%e7%ac%94%e8%ae%b0%ef%bc%885%ef%bc%89","status":"publish","type":"post","link":"https:\/\/www.gislxz.com\/index.php\/2022\/08\/01\/%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%e7%ac%94%e8%ae%b0%ef%bc%885%ef%bc%89\/","title":{"rendered":"\u6df1\u5ea6\u5b66\u4e60\u7b14\u8bb0\uff085\uff09"},"content":{"rendered":"\n<p>\u4e4b\u524d\u7684\u6a21\u578b\u8fd8\u662f\u57fa\u4e8e\u5355\u5c42\u7ebf\u6027\u56de\u5f52\u7684\u7f51\u7edc\uff0c\u5bf9mnist\u624b\u5199\u6570\u5b57\u7684\u8bc6\u522b\u6548\u679c\u5e76\u4e0d\u597d\uff0c\u8fd9\u4e00\u7ae0\u6765\u8ddf\u968f<a href=\"https:\/\/aistudio.baidu.com\/aistudio\/projectdetail\/1590916\" target=\"_blank\"  rel=\"nofollow\" >paddle\u5b98\u65b9\u6559\u7a0b<\/a>\u6765\u5b9e\u73b0\u591a\u5c42\u611f\u77e5\u7f51\u7edc\u548c\u5377\u79ef\u7f51\u7edc\u3002<\/p>\n\n\n\n<p>\u6570\u636e\u5904\u7406\u548c\u52a0\u8f7d\u51fd\u6570\u548c\u4e0a\u4e00\u7ae0\u4e00\u6837<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>#\u6570\u636e\u5904\u7406\u90e8\u5206\u4e4b\u524d\u7684\u4ee3\u7801\uff0c\u4fdd\u6301\u4e0d\u53d8\nimport os\nimport random\nimport paddle\nimport numpy as np\nimport matplotlib.pyplot as plt\nfrom PIL import Image\n\nimport gzip\nimport json\n\n# \u5b9a\u4e49\u6570\u636e\u96c6\u8bfb\u53d6\u5668\ndef load_data(mode='train'):\n\n    # \u52a0\u8f7d\u6570\u636e\n    datafile = '.\/work\/mnist.json.gz'\n    print('loading mnist dataset from {} ......'.format(datafile))\n    data = json.load(gzip.open(datafile))\n    print('mnist dataset load done')\n\n    # \u8bfb\u53d6\u5230\u7684\u6570\u636e\u533a\u5206\u8bad\u7ec3\u96c6\uff0c\u9a8c\u8bc1\u96c6\uff0c\u6d4b\u8bd5\u96c6\n    train_set, val_set, eval_set = data\n\n    # \u6570\u636e\u96c6\u76f8\u5173\u53c2\u6570\uff0c\u56fe\u7247\u9ad8\u5ea6IMG_ROWS, \u56fe\u7247\u5bbd\u5ea6IMG_COLS\n    # IMG_ROWS = 28\n    # IMG_COLS = 28\n\n    if mode == 'train':\n        # \u83b7\u5f97\u8bad\u7ec3\u6570\u636e\u96c6\n        imgs, labels = train_set&#91;0], train_set&#91;1]\n    elif mode == 'valid':\n        # \u83b7\u5f97\u9a8c\u8bc1\u6570\u636e\u96c6\n        imgs, labels = val_set&#91;0], val_set&#91;1]\n    elif mode == 'eval':\n        # \u83b7\u5f97\u6d4b\u8bd5\u6570\u636e\u96c6\n        imgs, labels = eval_set&#91;0], eval_set&#91;1]\n    else:\n        raise Exception(\"mode can only be one of &#91;'train', 'valid', 'eval']\")\n\n    #\u6821\u9a8c\u6570\u636e\n    imgs_length = len(imgs)\n    assert len(imgs) == len(labels), \\\n          \"length of train_imgs({}) should be the same as train_labels({})\".format(\n                  len(imgs), len(labels))\n\n    # \u5b9a\u4e49\u6570\u636e\u96c6\u6bcf\u4e2a\u6570\u636e\u7684\u5e8f\u53f7\uff0c \u6839\u636e\u5e8f\u53f7\u8bfb\u53d6\u6570\u636e\n    index_list = list(range(imgs_length))\n    # \u8bfb\u5165\u6570\u636e\u65f6\u7528\u5230\u7684batchsize\n    BATCHSIZE = 100\n\n    # \u5b9a\u4e49\u6570\u636e\u751f\u6210\u5668\n    def data_generator():\n        if mode == 'train':\n            random.shuffle(index_list)\n        imgs_list = &#91;]\n        labels_list = &#91;]\n        for i in index_list:\n            img = np.array(imgs&#91;i]).astype('float32')\n            label = np.array(labels&#91;i]).astype('float32')\n            # img = np.reshape(imgs&#91;i], &#91;1, IMG_ROWS, IMG_COLS]).astype('float32')\n            # label = np.reshape(labels&#91;i], &#91;1]).astype('float32')\n            imgs_list.append(img) \n            labels_list.append(label)\n            if len(imgs_list) == BATCHSIZE:\n                yield np.array(imgs_list), np.array(labels_list)\n                imgs_list = &#91;]\n                labels_list = &#91;]\n\n        # \u5982\u679c\u5269\u4f59\u6570\u636e\u7684\u6570\u76ee\u5c0f\u4e8eBATCHSIZE\uff0c\n        # \u5219\u5269\u4f59\u6570\u636e\u4e00\u8d77\u6784\u6210\u4e00\u4e2a\u5927\u5c0f\u4e3alen(imgs_list)\u7684mini-batch\n        if len(imgs_list) &gt; 0:\n            yield np.array(imgs_list), np.array(labels_list)\n\n    return data_generator<\/code><\/pre>\n\n\n\n<p>\u63a5\u7740\u8ddf\u7740\u6559\u7a0b\u5b9e\u73b0\u4e00\u4e0b\u56db\u5c42\u7f51\u7edc\uff0c\u6fc0\u6d3b\u51fd\u6570\u7528sigmoid<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/ai-studio-static-online.cdn.bcebos.com\/2173259df0704335b230ec158be0427677b9c77fd42348a28f2f8adf1ac1c706\" alt=\"\"\/><\/figure>\n\n\n\n<pre class=\"wp-block-code\"><code>import paddle.nn.functional as F\nfrom paddle.nn import Linear\n\n# \u5b9a\u4e49\u591a\u5c42\u5168\u8fde\u63a5\u795e\u7ecf\u7f51\u7edc\nclass MNIST(paddle.nn.Layer):\n    def __init__(self):\n        super(MNIST, self).__init__()\n        # \u5b9a\u4e49\u4e24\u5c42\u5168\u8fde\u63a5\u9690\u542b\u5c42\uff0c\u8f93\u51fa\u7ef4\u5ea6\u662f10\uff0c\u5f53\u524d\u8bbe\u5b9a\u9690\u542b\u8282\u70b9\u6570\u4e3a10\uff0c\u53ef\u6839\u636e\u4efb\u52a1\u8c03\u6574\n        self.fc1 = Linear(in_features=784, out_features=10)\n        self.fc2 = Linear(in_features=10, out_features=10)\n        # \u5b9a\u4e49\u4e00\u5c42\u5168\u8fde\u63a5\u8f93\u51fa\u5c42\uff0c\u8f93\u51fa\u7ef4\u5ea6\u662f1\n        self.fc3 = Linear(in_features=10, out_features=1)\n    \n    # \u5b9a\u4e49\u7f51\u7edc\u7684\u524d\u5411\u8ba1\u7b97\uff0c\u9690\u542b\u5c42\u6fc0\u6d3b\u51fd\u6570\u4e3asigmoid\uff0c\u8f93\u51fa\u5c42\u4e0d\u4f7f\u7528\u6fc0\u6d3b\u51fd\u6570\n    def forward(self, inputs):\n        # inputs = paddle.reshape(inputs, &#91;inputs.shape&#91;0], 784])\n        outputs1 = self.fc1(inputs)\n        outputs1 = F.sigmoid(outputs1)\n        outputs2 = self.fc2(outputs1)\n        outputs2 = F.sigmoid(outputs2)\n        outputs_final = self.fc3(outputs2)\n        return outputs_final<\/code><\/pre>\n\n\n\n<p>\u4e4b\u540e\u6765\u5199\u8bad\u7ec3\u8fc7\u7a0b\uff0c\u6362\u6210SGD\u4f18\u5316\u5668\uff0cpaddle\u8fd9SGD\u4f18\u5316\u5668\u7684\u8bed\u6cd5\u5012\u662f\u548cpytorch\u4e00\u6837\u4e86<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>#\u7f51\u7edc\u7ed3\u6784\u90e8\u5206\u4e4b\u540e\u7684\u4ee3\u7801\uff0c\u4fdd\u6301\u4e0d\u53d8\ndef train(model):\n    model.train()\n    #\u8c03\u7528\u52a0\u8f7d\u6570\u636e\u7684\u51fd\u6570\uff0c\u83b7\u5f97MNIST\u8bad\u7ec3\u6570\u636e\u96c6\n    train_loader = load_data('train')\n    # \u4f7f\u7528SGD\u4f18\u5316\u5668\uff0clearning_rate\u8bbe\u7f6e\u4e3a0.01\n    opt = paddle.optimizer.SGD(learning_rate=0.01, parameters=model.parameters())\n    # \u8bad\u7ec35\u8f6e\n    EPOCH_NUM = 5\n    for epoch_id in range(EPOCH_NUM):\n        for batch_id, data in enumerate(train_loader()):\n            #\u51c6\u5907\u6570\u636e\n            images, labels = data\n            images = paddle.to_tensor(images)\n            labels = paddle.to_tensor(labels)\n            \n            #\u524d\u5411\u8ba1\u7b97\u7684\u8fc7\u7a0b\n            predicts = model(images)\n            \n            #\u8ba1\u7b97\u635f\u5931\uff0c\u53d6\u4e00\u4e2a\u6279\u6b21\u6837\u672c\u635f\u5931\u7684\u5e73\u5747\u503c\n            loss = F.square_error_cost(predicts, labels)\n            avg_loss = paddle.mean(loss)\n\n            #\u6bcf\u8bad\u7ec3200\u6279\u6b21\u7684\u6570\u636e\uff0c\u6253\u5370\u4e0b\u5f53\u524dLoss\u7684\u60c5\u51b5\n            if batch_id % 200 == 0:\n                print(\"epoch: {}, batch: {}, loss is: {}\".format(epoch_id, batch_id, avg_loss.numpy()))\n            \n            #\u540e\u5411\u4f20\u64ad\uff0c\u66f4\u65b0\u53c2\u6570\u7684\u8fc7\u7a0b\n            avg_loss.backward()\n            # \u6700\u5c0f\u5316loss,\u66f4\u65b0\u53c2\u6570\n            opt.step()\n            # \u6e05\u9664\u68af\u5ea6\n            opt.clear_grad()\n\n    #\u4fdd\u5b58\u6a21\u578b\u53c2\u6570\n    paddle.save(model.state_dict(), 'mnist.pdparams')\n\nmodel = MNIST()\ntrain(model)<\/code><\/pre>\n\n\n\n<p>\u6309\u9053\u7406\u8fd9\u79cd\u5206\u7c7b\u95ee\u9898\u5e94\u8be5\u662f\u8f93\u51fa10\u4e2a\u7ed3\u679c\uff0c\u518d\u7528softmax\u51fd\u6570\u5f97\u5230\u6700\u540e\u7ed3\u679c\u7684\uff0c\u4e0d\u8fc7\u8fd9\u4e00\u7ae0\u6559\u7a0b\u4e3b\u8981\u7814\u7a76\u7f51\u7edc\u7ed3\u6784\uff0c\u635f\u5931\u51fd\u6570\u4ee5\u53casoftmax\u51fd\u6570\u5728\u4e0b\u4e00\u7ae0\u6559\u7a0b\u8bb2\u3002<\/p>\n\n\n\n<p>\u6211\u4eec\u7528torch\u4e5f\u6765\u5b9e\u73b0\u4e00\u4e0b,\u6539\u4e00\u4e0bMnist\u7c7b\u7684\u7ed3\u6784\u5c31\u884c\u4e86\uff0c\u5176\u4ed6\u90fd\u4e0d\u53d8\uff0c\u672c\u6765torch\u7684\u6fc0\u6d3b\u51fd\u6570\u548cpaddle\u662f\u4e00\u4e2a\u8def\u5f84\u7684\uff0c\u5373torch.nn.functional\uff0c\u4f46\u73b0\u5728\u65b0\u7684\u7248\u672c\u63d0\u793a\u6211\u7528torch.sigmoid\u5c31\u597d<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>import torch\nfrom torch import nn\nfrom torch import optim\nimport torch.nn.functional as F\nimport torchvision\nimport json\nimport numpy as np\nimport gzip\nimport random\n\nclass Mnist(nn.Module):\n    def __init__(self):\n        super(Mnist,self).__init__()\n        self.fc1 = nn.Linear(28*28*1,10)\n        self.fc2 = nn.Linear(10,10)\n        self.fc3 = nn.Linear(10,1)\n    def forward(self,x):\n        x = x.view(-1,28*28*1)\n        x = self.fc1(x)\n        x = torch.sigmoid(x)\n        x = self.fc2(x)\n        x = torch.sigmoid(x)\n        x = self.fc3(x)\n        x = torch.sigmoid(x)\n        return x\n\n# \u8bad\u7ec3\u914d\u7f6e\uff0c\u5e76\u542f\u52a8\u8bad\u7ec3\u8fc7\u7a0b\nmodel = Mnist()\nmodel.train(mode=True)\n#\u8c03\u7528\u52a0\u8f7d\u6570\u636e\u7684\u51fd\u6570\ntrain_loader = load_data('train')\noptimizer = optim.SGD(model.parameters(),lr= 0.001)\ncriterion = nn.MSELoss()\n\nEPOCH_NUM = 10\nfor epoch_id in range(EPOCH_NUM):\n    for batch_id, data in enumerate(train_loader()):\n        #\u51c6\u5907\u6570\u636e\uff0c\u53d8\u5f97\u66f4\u52a0\u7b80\u6d01\n        image_data, label_data = data\n        image = torch.tensor(image_data)\n        label = torch.tensor(label_data)\n            \n        #\u524d\u5411\u8ba1\u7b97\u7684\u8fc7\u7a0b\n        predict = model(image)\n            \n        #\u8ba1\u7b97\u635f\u5931\uff0c\u53d6\u4e00\u4e2a\u6279\u6b21\u6837\u672c\u635f\u5931\u7684\u5e73\u5747\u503c\n        loss = criterion(predict, label)\n        avg_loss = torch.mean(loss)\n            \n        #\u6bcf\u8bad\u7ec3\u4e86200\u6279\u6b21\u7684\u6570\u636e\uff0c\u6253\u5370\u4e0b\u5f53\u524dLoss\u7684\u60c5\u51b5\n        if batch_id % 200 == 0:\n            print(\"epoch: {}, batch: {}, loss is: {}\".format(epoch_id, batch_id, avg_loss.detach().numpy()))\n            \n        #\u540e\u5411\u4f20\u64ad\uff0c\u66f4\u65b0\u53c2\u6570\u7684\u8fc7\u7a0b\n        avg_loss.backward()\n        optimizer.step()\n        model.zero_grad()<\/code><\/pre>\n\n\n\n<p>\u8fd9loss\u8fd8\u662f\u5f88\u5927\u554a\uff0c\u6ca1\u5565\u7528\uff0c\u56fe\u50cf\u5206\u7c7b\u8fd8\u662f\u5f97\u7528\u5377\u79ef\uff0cpaddle\u8fd9\u4e00\u7ae0\u6559\u7a0b\u8fd8\u6ca1\u6709\u7ec6\u8bf4\u5377\u79ef\u7684\u539f\u7406\uff0c\u4e0d\u8fc7\u6211\u4eec\u53ef\u4ee5\u5148\u6765\u5b9e\u73b0\u4e00\u4e0b\uff0c\u5177\u4f53\u539f\u7406\u53ef\u4ee5\u770b\u4e0b\u4e00\u7ae0\u6559\u7a0b\u3002\u9996\u5148\u6539\u4e00\u4e0b\u7f51\u7edc\u7ed3\u6784\uff0c\u6ce8\u610f\u5230\u8fd9\u91cc\u8f93\u5165\u7684\u662f\u5355\u901a\u9053\u56fe\u50cf\uff0c\u4e5f\u5c31\u662f\u30101\u00d728\u00d728\u3011\u7684\u6570\u7ec4\uff0c\u6240\u4ee5\u540e\u9762\u8f93\u5165\u6570\u636e\u4e5f\u8981reshape\u6210[batch_size, 1, 28, 28]<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code># \u5b9a\u4e49 SimpleNet \u7f51\u7edc\u7ed3\u6784\nimport paddle\nfrom paddle.nn import Conv2D, MaxPool2D, Linear\nimport paddle.nn.functional as F\n# \u591a\u5c42\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u5b9e\u73b0\nclass MNIST(paddle.nn.Layer):\n     def __init__(self):\n         super(MNIST, self).__init__()\n         \n         # \u5b9a\u4e49\u5377\u79ef\u5c42\uff0c\u8f93\u51fa\u7279\u5f81\u901a\u9053out_channels\u8bbe\u7f6e\u4e3a20\uff0c\u5377\u79ef\u6838\u7684\u5927\u5c0fkernel_size\u4e3a5\uff0c\u5377\u79ef\u6b65\u957fstride=1\uff0cpadding=2\n         self.conv1 = Conv2D(in_channels=1, out_channels=20, kernel_size=5, stride=1, padding=2)\n         # \u5b9a\u4e49\u6c60\u5316\u5c42\uff0c\u6c60\u5316\u6838\u7684\u5927\u5c0fkernel_size\u4e3a2\uff0c\u6c60\u5316\u6b65\u957f\u4e3a2\n         self.max_pool1 = MaxPool2D(kernel_size=2, stride=2)\n         # \u5b9a\u4e49\u5377\u79ef\u5c42\uff0c\u8f93\u51fa\u7279\u5f81\u901a\u9053out_channels\u8bbe\u7f6e\u4e3a20\uff0c\u5377\u79ef\u6838\u7684\u5927\u5c0fkernel_size\u4e3a5\uff0c\u5377\u79ef\u6b65\u957fstride=1\uff0cpadding=2\n         self.conv2 = Conv2D(in_channels=20, out_channels=20, kernel_size=5, stride=1, padding=2)\n         # \u5b9a\u4e49\u6c60\u5316\u5c42\uff0c\u6c60\u5316\u6838\u7684\u5927\u5c0fkernel_size\u4e3a2\uff0c\u6c60\u5316\u6b65\u957f\u4e3a2\n         self.max_pool2 = MaxPool2D(kernel_size=2, stride=2)\n         # \u5b9a\u4e49\u4e00\u5c42\u5168\u8fde\u63a5\u5c42\uff0c\u8f93\u51fa\u7ef4\u5ea6\u662f1\n         self.fc = Linear(in_features=980, out_features=1)\n         \n    # \u5b9a\u4e49\u7f51\u7edc\u524d\u5411\u8ba1\u7b97\u8fc7\u7a0b\uff0c\u5377\u79ef\u540e\u7d27\u63a5\u7740\u4f7f\u7528\u6c60\u5316\u5c42\uff0c\u6700\u540e\u4f7f\u7528\u5168\u8fde\u63a5\u5c42\u8ba1\u7b97\u6700\u7ec8\u8f93\u51fa\n    # \u5377\u79ef\u5c42\u6fc0\u6d3b\u51fd\u6570\u4f7f\u7528Relu\uff0c\u5168\u8fde\u63a5\u5c42\u4e0d\u4f7f\u7528\u6fc0\u6d3b\u51fd\u6570\n     def forward(self, inputs):\n         x = self.conv1(inputs)\n         x = F.relu(x)\n         x = self.max_pool1(x)\n         x = self.conv2(x)\n         x = F.relu(x)\n         x = self.max_pool2(x)\n         x = paddle.reshape(x, &#91;x.shape&#91;0], -1])\n         x = self.fc(x)\n         return x<\/code><\/pre>\n\n\n\n<p>\u8bad\u7ec3\u8fc7\u7a0b\u4e5f\u8981\u6539\u4e00\u4e0b\u8f93\u5165shape<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>#\u7f51\u7edc\u7ed3\u6784\u90e8\u5206\u4e4b\u540e\u7684\u4ee3\u7801\uff0c\u4fdd\u6301\u4e0d\u53d8\ndef train(model):\n    model.train()\n    #\u8c03\u7528\u52a0\u8f7d\u6570\u636e\u7684\u51fd\u6570\uff0c\u83b7\u5f97MNIST\u8bad\u7ec3\u6570\u636e\u96c6\n    train_loader = load_data('train')\n    # \u4f7f\u7528SGD\u4f18\u5316\u5668\uff0clearning_rate\u8bbe\u7f6e\u4e3a0.01\n    opt = paddle.optimizer.SGD(learning_rate=0.01, parameters=model.parameters())\n\n    # \u8bad\u7ec35\u8f6e\n    EPOCH_NUM = 5\n    # MNIST\u56fe\u50cf\u9ad8\u548c\u5bbd\n    IMG_ROWS, IMG_COLS = 28, 28\n\n    for epoch_id in range(EPOCH_NUM):\n        for batch_id, data in enumerate(train_loader()):\n            #\u51c6\u5907\u6570\u636e\n            images, labels = data\n            images = paddle.to_tensor(images)\n            images = paddle.reshape(images, &#91;images.shape&#91;0],1,IMG_ROWS,IMG_COLS])\n            labels = paddle.to_tensor(labels)\n            \n            #\u524d\u5411\u8ba1\u7b97\u7684\u8fc7\u7a0b\n            predicts = model(images)\n            \n            #\u8ba1\u7b97\u635f\u5931\uff0c\u53d6\u4e00\u4e2a\u6279\u6b21\u6837\u672c\u635f\u5931\u7684\u5e73\u5747\u503c\n            loss = F.square_error_cost(predicts, labels)\n            avg_loss = paddle.mean(loss)\n            \n            #\u6bcf\u8bad\u7ec3\u4e86200\u6279\u6b21\u7684\u6570\u636e\uff0c\u6253\u5370\u4e0b\u5f53\u524dLoss\u7684\u60c5\u51b5\n            if batch_id % 200 == 0:\n                print(\"epoch: {}, batch: {}, loss is: {}\".format(epoch_id, batch_id, avg_loss.numpy()))\n            \n            #\u540e\u5411\u4f20\u64ad\uff0c\u66f4\u65b0\u53c2\u6570\u7684\u8fc7\u7a0b\n            avg_loss.backward()\n            # \u6700\u5c0f\u5316loss,\u66f4\u65b0\u53c2\u6570\n            opt.step()\n            # \u6e05\u9664\u68af\u5ea6\n            opt.clear_grad()\n\n    #\u4fdd\u5b58\u6a21\u578b\u53c2\u6570\n    paddle.save(model.state_dict(), 'mnist.pdparams')\n\n#\u521b\u5efa\u6a21\u578b    \nmodel = MNIST()\n#\u542f\u52a8\u8bad\u7ec3\u8fc7\u7a0b\ntrain(model)<\/code><\/pre>\n\n\n\n<p>torch\u7248\u672c\u6211\u4eec\u4e5f\u6539\u4e00\u4e0b<\/p>\n\n\n\n<pre class=\"wp-block-code\"><code>from torch.nn import Conv2d,MaxPool2d,Linear\n#\u6570\u636e\u5904\u7406\u90e8\u5206\u4e4b\u540e\u7684\u4ee3\u7801\uff0c\u6570\u636e\u8bfb\u53d6\u7684\u90e8\u5206\u8c03\u7528load_data\u51fd\u6570\n# \u5b9a\u4e49\u7f51\u7edc\u7ed3\u6784\uff0c\u540c\u4e0a\u4e00\u8282\u6240\u4f7f\u7528\u7684\u7f51\u7edc\u7ed3\u6784\nclass Mnist(nn.Module):\n    def __init__(self):\n        super(Mnist,self).__init__()\n        self.conv1 = Conv2d(in_channels=1,out_channels=20,kernel_size=5,stride=1,padding=2)\n        self.max_pool1 = MaxPool2d(kernel_size=2,stride=2)\n        self.conv2 = Conv2d(in_channels=20,out_channels=20,kernel_size=5,stride=1,padding=2)\n        self.max_pool2 = MaxPool2d(kernel_size=2,stride=2)\n        self.fc = Linear(in_features=980,out_features=1)\n    def forward(self,x):\n        x = self.conv1(x)\n        x = torch.relu(x)\n        x = self.max_pool1(x)\n        x = self.conv2(x)\n        x = torch.relu(x)\n        x = self.max_pool2(x)\n        x = torch.reshape(x,&#91;x.shape&#91;0],-1])\n        x = self.fc(x)\n        return x\n\n# \u8bad\u7ec3\u914d\u7f6e\uff0c\u5e76\u542f\u52a8\u8bad\u7ec3\u8fc7\u7a0b\nmodel = Mnist()\nmodel.train(mode=True)\n#\u8c03\u7528\u52a0\u8f7d\u6570\u636e\u7684\u51fd\u6570\ntrain_loader = load_data('train')\noptimizer = optim.SGD(model.parameters(),lr= 0.001)\ncriterion = nn.MSELoss()\n\nEPOCH_NUM = 10\nfor epoch_id in range(EPOCH_NUM):\n    for batch_id, data in enumerate(train_loader()):\n        #\u51c6\u5907\u6570\u636e\uff0c\u53d8\u5f97\u66f4\u52a0\u7b80\u6d01\n        image_data, label_data = data\n        image = torch.tensor(image_data)\n        label = torch.tensor(label_data)\n        image = torch.reshape(image,&#91;image.shape&#91;0],1,28,28]) \n        #\u524d\u5411\u8ba1\u7b97\u7684\u8fc7\u7a0b\n        predict = model(image)\n            \n        #\u8ba1\u7b97\u635f\u5931\uff0c\u53d6\u4e00\u4e2a\u6279\u6b21\u6837\u672c\u635f\u5931\u7684\u5e73\u5747\u503c\n        loss = criterion(predict, label)\n        avg_loss = torch.mean(loss)\n            \n        #\u6bcf\u8bad\u7ec3\u4e86200\u6279\u6b21\u7684\u6570\u636e\uff0c\u6253\u5370\u4e0b\u5f53\u524dLoss\u7684\u60c5\u51b5\n        if batch_id % 200 == 0:\n            print(\"epoch: {}, batch: {}, loss is: {}\".format(epoch_id, batch_id, avg_loss.detach().numpy()))\n            \n        #\u540e\u5411\u4f20\u64ad\uff0c\u66f4\u65b0\u53c2\u6570\u7684\u8fc7\u7a0b\n        avg_loss.backward()\n        optimizer.step()\n        model.zero_grad()<\/code><\/pre>\n\n\n\n<p>\u5377\u79ef\u7f51\u7edcloss\u5c31\u5c0f\u4e86\u5f88\u591a\u4e86<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"604\" height=\"675\" src=\"http:\/\/www.gislxz.com\/wp-content\/uploads\/2022\/07\/7748390545BFF221F76BB5D840C70D4C.jpg\" alt=\"\" class=\"wp-image-353\" srcset=\"https:\/\/www.gislxz.com\/wp-content\/uploads\/2022\/07\/7748390545BFF221F76BB5D840C70D4C.jpg 604w, https:\/\/www.gislxz.com\/wp-content\/uploads\/2022\/07\/7748390545BFF221F76BB5D840C70D4C-268x300.jpg 268w\" sizes=\"auto, (max-width: 604px) 100vw, 604px\" \/><\/figure>\n","protected":false},"excerpt":{"rendered":"<p>mnist\u624b\u5199\u6570\u5b57\u8bc6\u522b\u4e4b\u7f51\u7edc\u7ed3\u6784\uff08\u591a\u5c42\u611f\u77e5\/\u5377\u79ef\uff09\u2014\u2014paddle\u548cpytorch\u5bf9\u6bd4\u5b9e\u73b0<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[21],"tags":[],"class_list":["post-383","post","type-post","status-publish","format-standard","hentry","category-21"],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/posts\/383","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/comments?post=383"}],"version-history":[{"count":2,"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/posts\/383\/revisions"}],"predecessor-version":[{"id":387,"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/posts\/383\/revisions\/387"}],"wp:attachment":[{"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/media?parent=383"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/categories?post=383"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.gislxz.com\/index.php\/wp-json\/wp\/v2\/tags?post=383"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}